Thanks for making this great toolbox!
I have a question about negative values:
When converting a single image into cone capture rates there is a tick box allowing all negative values to be converted to 0.001.
However, I cannot find this option for when running the Batch Multispectral Image Analysis. This means I can get different outputs depending on whether an image is run individually or as part of a batch.
Is there an option or stage that I am missing?
Negative values aren’t actually an issue for the bandpass-based methods (they can deal with zero or negative values happily, whether FFT or Gaussian convolution). We need to remove negative values for the RNL model due to divide-by-zero infinity issues.
So I’d never thought about this issue… There’s also the question of why the negative values are created (can either mean your grey standards or black-point are slightly off, or that the colour is out-of-gamut for the camera or visual system). In the former case (which will be the most common cause) it’s better to actually keep negative values because they’re informative, rather than impossible.
Anyway, it should be straightforward to add the option if you think it’s necessary?
We’re using the batch processing for RNL modelling rather than bandpass pattern analysis.
The discrepancy is happening occasionally where we have ROIs that inclusively cover very dark regions (true blacks across VIS and UV). So I think a value very close to zero is correct and negatives may result from noise within the camera sensor (old Canon 7d), does this sound reasonable?
The RNL model seemingly works perfectly when analysing an image individually and replacing the negative values.
If you are able to add the ‘remove negative values’ option to the ‘batch multispectral image analysis’ I think that would help with consistency between the two processes.